Development and Validation of a Clinically Actionable Prediction Model for Postoperative Pulmonary Complications in Cardiac Surgery: A Focus on Modifiable Risk Factors
1Cardiopulmonary Rehabilitation Center, National Center for Cardiovascular Disease, China &Fuwai Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China
2Department of Surgery, National Center for Cardiovascular Disease, China &Fuwai Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China
3National Clinical Research Center for Cardiovascular Diseases, China &Fuwai Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China
4Center of Cardiac Surgery in Adults, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China
5Information Center, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China
Correspondence: Ruoxi Li Cardiopulmonary Rehabilitation Center, National Center for Cardiovascular Disease, China & Fuwai Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, A 167, Beilishi Road, Xicheng District, Beijing, 100037, China. Tel: +86-10-88396212 Fax: +86-10-68314466 E-mail: liruoxi2025sci@163.com
*These authors contributed equally to this work.
• Received: June 30, 2025 • Revised: October 9, 2025 • Accepted: December 12, 2025
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
To develop and validate a clinically actionable prediction model for postoperative pulmonary complications (PPCs) in cardiac surgery patients, focusing on modifiable preoperative risk factors amenable to targeted optimization.
Methods
In this prospective observational cohort study, 492 adults undergoing open-chest cardiac surgery between August 15, 2023 and December 31, 2023 were analyzed. Prespecified predictors included gas exchange variables, pulmonary function, inspiratory muscle strength, and physical performance. Univariable and multivariable logistic regression analyses were used to develop the prediction model. Discrimination was assessed by the area under the receiver operating characteristic curve (AUC).
Results
A total of 90 patients (14.1%) developed PPCs after surgery. Five independent predictors were identified: elevated arterial PaCO2 (odds ratio [OR] 1.12, 95% confidence interval [CI] 1.00–1.26), oxygen desaturation (SpO2<93%) (OR 12.47, 95% CI 3.51–48.13), reduced gait speed (OR 0.17, 95% CI 0.04–0.71), lower FEV1/FVC ratio (OR 0.96, 95% CI 0.92–1.00), and diminished inspiratory muscle strength (MIP % predicted) (OR 0.96, 95% CI 0.92–0.99). The model demonstrated good discriminative ability with an AUC of 0.86 (95% CI 0.80–0.93) in the training cohort and 0.87 (95% CI 0.74–0.93) in the validation cohort.
Conclusion
This parsimonious model achieved high predictive accuracy using five modifiable physiological variables. By targeting abnormalities in gas exchange, pulmonary mechanics, muscle strength, and functional reserve, the model offers a practical tool to guide individualized prehabilitation strategies for reducing PPC risk in cardiac surgery patients.
Postoperative pulmonary complications (PPCs) are among the most frequent and severe complications following cardiac surgery, including reintubation, prolonged mechanical ventilation (>24 hours), tracheostomy, and pneumonia [1-3]. The incidence of PPCs varies widely (3%–50%), depending on definitions and surgical techniques [4], and their occurrence significantly increases morbidity, length of hospital stay, and healthcare costs [5-8]. Given these consequences, early identification of high-risk patients is critical for optimizing perioperative care and guiding targeted interventions.
Several predictive models have been developed to assess PPC risk [2,5,9-11], but they remain limited in clinical applicability. Many existing models are derived from heterogeneous surgical populations, primarily focusing on general anesthesia rather than being specific to cardiac surgery [2,5,9-11]. Additionally, they tend to emphasize clinical and demographic characteristics while overlooking modifiable preoperative factors such as respiratory muscle strength, functional capacity, and nutritional status. Furthermore, most models incorporate a large number of predictors, sometimes exceeding ten, which reduces their practicality for real-time decision-making. Finally, a crucial gap in current research is the lack of integration of prehabilitation and rehabilitation factors, despite growing evidence that optimizing respiratory function and muscle strength before surgery can significantly mitigate PPC risk.
Cardiac rehabilitation and prehabilitation play a pivotal role in improving postoperative outcomes, yet they are rarely considered in PPC prediction models. Preoperative inspiratory muscle training (IMT), structured physical activity, and nutritional optimization have been shown to enhance respiratory function, functional reserve, and perioperative resilience. Since respiratory impairment and muscle deconditioning are prevalent among cardiac surgery patients, incorporating these modifiable factors into predictive models is essential for improving risk stratification and guiding personalized interventions.
Given these gaps, this study aimed to develop and validate a clinically actionable prediction model for PPCs in cardiac surgery patients, focusing on modifiable risk factors related to respiratory function, muscle strength, and preoperative rehabilitation. By integrating functional and biochemical predictors, our model seeks to enhance early risk assessment and enable targeted preoperative interventions to reduce PPC incidence and improve postoperative recovery.
METHODS
Study design
This prospective, single-center observational study was conducted at Fuwai Hospital, Chinese Academy of Medical Sciences from August 15, 2023 to December 31, 2023, following established methodologies for predictive model development [11]. A prospective design was chosen to minimize recall bias and enhance the accuracy of perioperative risk factor assessment. Patients were randomly assigned to development and validation cohorts in a 7:3 ratio using stratified randomization based on key baseline characteristics.
Patients
Patients aged ≥18 years scheduled for elective open-chest cardiac surgery between between August 15, 2023 and December 31, 2023, were eligible for inclusion. Exclusion criteria included: (1) emergency surgery; (2) preoperative pulmonary infection confirmed by positive sputum culture; (3) history of prior pulmonary surgery; (4) severe cognitive impairment or psychiatric disorders precluding cooperation; and (5) coma or heavy sedation at admission. The study was approved by the Institutional Review Board of Fuwai Hospital, Chinese Academy of Medical Sciences (approval no. 2023-2108), and written informed consent was obtained from all participants.
Cardiac operations were categorized a priori from the operative record into five mutually exclusive groups: (1) coronary artery bypass grafting (CABG)—on-pump and off-pump CABG analyzed together; (2) isolated valve surgery—aortic, mitral, or tricuspid valve procedures (single or multiple valves); (3) CABG+valve—combined coronary revascularization and valve surgery during the same operation; (4) aortic surgery—ascending and/or arch procedures not meeting the combined category; and (5) other cardiac procedures—less common open-chest operations not classified above (e.g., atrial septal defect closure, ventricular septal defect patch repair, atrial myxoma excision, patent ductus arteriosus ligation, pericardiectomy).
Baseline respiratory disease was captured as chronic obstructive pulmonary disease (COPD) based on the preoperative medical record and/or spirometry consistent with airflow obstruction (post-bronchodilator forced expiratory volume in 1 second/forced vital capacity [FEV1/FVC]<0.70 when available). By institutional policy for elective cardiac surgery, patients with severe pulmonary conditions unlikely to tolerate surgery—such as advanced interstitial lung disease, active pulmonary infection, or end-stage lung malignancy—were not enrolled; consequently, these diagnoses are absent from the cohort and are not tabulated.
Potential predictors
Potential predictive variables were selected a priori based on prior literature, physiological plausibility, and clinical relevance [12,13]. These variables were categorized into three groups:
1. Demographic and lifestyle factors: age, sex, body mass index (BMI), smoking history, diabetes mellitus.
2. Laboratory biomarkers: pre-albumin, total protein, albumin, low-density lipoprotein cholesterol, ferritin, arterial blood gas parameters (partial pressure of arterial carbon dioxide [PaCO2], partial pressure of arterial oxygen [PaO2], arterial oxygen saturation [SaO2]), and anemia.
3. Functional and respiratory assessments: grip strength index, gait speed, chair rise time, pulmonary function parameters (FVC, peak expiratory flow, FEV1, FEV1/FVC ratio, maximal voluntary ventilation), and respiratory muscle strength examination (maximal inspiratory pressure [MIP], MIP [% of predicted], maximal expiratory pressure [MEP], MEP [% of predicted]).
Preoperative peripheral capillary oxygen saturation (SpO2) was obtained on room air (FiO2=0.21) with the patient in the supine position after ≥5 minutes of quiet rest. Measurements were taken using hospital-grade bedside monitors equipped with pulse-oximetry modules and a finger probe. Readings were accepted only when the plethysmographic waveform was stable and pulse-synchronous; measurements with motion artifact, poor perfusion, or nail polish were repeated after correction. For modeling, desaturation was defined a priori as SpO2<93%.
All data were collected preoperatively and verified through a double-checking process before analysis.
Outcome definition
The primary outcome was the occurrence of clinically significant PPCs, defined as any grades 3 or 4 event on the Kroenke Score [13-16]. One trained assessor, blinded to all baseline predictors, subgroup assignments, and model outputs, abstracted and scored outcomes from the medical record and radiology reports. To pragmatically document reproducibility, a second blinded assessor independently re-scored a random 15% subsample (n=74), and disagreements were resolved by discussion.
Statistical analysis
A sample size calculation was performed based on an expected PPC incidence of 20% and an estimated risk ratio of 1.9–3.2 for modifiable factors [17-19]. Using α=0.05, power=0.80, and an effect size of 0.5, a minimum of 154 patients were required. Accounting for a 20% dropout rate, the final estimated sample size was 487.
Continuous variables were reported as mean±standard deviation (SD) or median with interquartile range, while categorical variables were expressed as frequencies and percentages. Differences between PPC and non-PPC groups were assessed using unpaired t-tests for continuous variables and χ2 tests for categorical variables.
For the binary PPC outcome (grade≥3), Cohen’s κ with 95% confidence intervals (CIs) (nonparametric bootstrap, 5,000 resamples) and raw percent agreement were reported; positive and negative agreement were summarized from the 2×2 table.
Univariate logistic regression was used to identify potential predictors, with variables meeting p<0.01 entered into a multivariable logistic regression model. Stepwise backward selection was performed to develop the final model. Before final model estimation, we quantified multicollinearity using variance inflation factors (VIFs). We considered VIF>5 as thresholds of concern. Collinearity metrics were computed in the development cohort to avoid information leakage. Because SpO2 (pulse oximetry) and PaO2 (arterial oxygen tension) are related but not interchangeable, we examined pairwise correlations and VIFs for PaCO2, PaO2, and SpO2. A priori, VIF>5 was considered concerning. For model parsimony and bedside applicability, the primary multivariable model used an SpO2-based desaturation indicator (e.g., SpO2<93%) to represent gas exchange and did not include PaO2 concurrently.
Model discrimination was evaluated via the area under the receiver operating characteristic curve (AUC), and decision curve analysis (DCA) was conducted to assess clinical utility. Beyond the fixed 70/30 split, internal validation within the development cohort used bootstrap (500 resamples) to estimate optimism in discrimination (AUC) and overall accuracy (Brier), yielding optimism-corrected metrics; and repeated stratified 5-fold×10 cross-validation to summarize AUC and Brier as mean±SD across folds. Out-of-fold predictions were used for cross-validated metrics. To support clinical use, we prespecified three probability thresholds on the development cohort: Youden’s J (balanced), a rule-out threshold targeting sensitivity≥0.80, and a rule-in threshold targeting specificity≥0.80. These cut-points were evaluated in the validation cohort (sensitivity, specificity, positive predictive value, negative predictive value), and risk strata (low <0.10; intermediate 0.10–0.19; high ≥0.19) were defined for bedside triage. Statistical significance was set at p<0.05. All analyses were conducted using R software (version 4.0.4; R Foundation), and a nomogram was created using the rms package.
To assess the stability and generalizability of the prediction model, subgroup analyses were conducted based on clinically relevant stratifications, including age (>65 years), BMI (<25 kg/m2), and presence of diabetes mellitus. Within each subgroup, the model’s performance was evaluated using the AUC, sensitivity, and specificity, along with 95% CIs for AUC.
RESULTS
Patient characteristics
Flowchart of the study is shown in Fig. 1. A total of 492 patients undergoing open-chest cardiac surgery between August 15, 2023 and December 31, 2023 were included. According to the Kroenke Score, 90 patients (14.1%) developed PPCs after the surgery. In the random subsample (n=74) independently re-scored by a second blinded assessor, agreement was 97.3%, with Cohen’s κ=0.893 (95% CI 0.705–1.000). Positive and negative agreement were 90.9% and 98.4%, respectively.
Baseline characteristics of patients with and without PPCs in the training cohort are summarized in Table 1. The procedures were mainly CABG (281 out of 492, 57.1%), followed by isolated valve surgery (126 out of 492, 25.6%), CABG+valve (29 out of 492, 5.9%), aortic surgery (9 out of 492, 1.8%), and other procedures (47 out of 492, 9.6%). The proportion of PPCs did not differ significantly across categories (global χ2 p=0.412; Supplementary Table S1).
Compared to the non-PPC group, patients who developed PPCs tended to be older (61.7±9.2 years vs. 59.0±11.1 years, p=0.11), although the difference was not statistically significant. The PPC group had significantly higher preoperative arterial carbon dioxide partial pressure (PaCO2: 41.9±5.0 mmHg vs. 38.3±3.5 mmHg, p<0.001), and lower arterial oxygen partial pressure (PaO2: 65.5±23.0 mmHg vs. 89.4±10.9 mmHg, p<0.001). Additionally, a markedly higher proportion of patients in the PPC group had oxygen saturation (SpO2)<93% (61.2% vs. 4.1%, p<0.001).
Functionally, patients in the PPC group demonstrated significantly slower gait speed (median 1.2 m/s vs. 1.3 m/s, p=0.02), lower inspiratory muscle strength as measured by MIP (% predicted) (63.2±19.2% vs. 69.6±15.9%, p=0.01), and reduced FEV1/FVC ratio (72.9±9.7% vs. 78.0±9.3%, p=0.01). These differences suggest that impaired gas exchange, decreased respiratory muscle performance, and reduced physical reserve may contribute to the development of PPCs in this population.
Additional baseline comparisons for the overall cohort and the validation cohort are presented in Supplementary Tables S2 and S3, respectively.
Model development and validation
Table 2 summarizes the results of the univariable and the multivariable logistic regression analyses. Five independent predictors of PPCs were identified using a backward stepwise variable selection method: elevated arterial PaCO2 (odds ratio [OR] 1.12, 95% CI 1.00–1.26), oxygen desaturation (SpO2<93%; OR 12.47, 95% CI 3.51–48.13), reduced gait speed (OR 0.17, 95% CI 0.04–0.71), lower FEV1/FVC ratio (OR 0.96, 95% CI 0.92–1.00), and diminished inspiratory muscle strength (MIP % predicted; OR 0.96, 95% CI 0.92–0.99). Among gas-exchange variables, PaO2 and SpO2 were moderately correlated (r=0.47), with low VIFs for PaCO2 (1.20), PaO2 (1.28), and SpO2 (1.47). Accordingly, the primary multivariable model included the SpO2-based desaturation indicator and excluded PaO2 (Supplementary Table S4). In the final multivariable model, VIFs were 1.29 (PaCO2), 1.30 (SpO2<93%), 1.00 (gait speed), 1.01 (FEV1/FVC%), and 1.00 (MIP % predicted), indicating no meaningful multicollinearity.
A multivariable prediction model incorporating these five variables was developed and demonstrated good discriminative performance, with an AUC of 0.86 (95% CI 0.80–0.93) in the training cohort and 0.87 (95% CI 0.74–0.93) in the validation cohort (Fig. 2). In the training cohort (n=345; events=49), bootstrap validation (500 resamples) yielded an optimism-corrected AUC of 0.667 (apparent 0.698, optimism 0.031) and an optimism-corrected Brier score of 0.119 (apparent 0.115). Repeated 5-fold×10 cross-validation produced AUC 0.658±0.079 and Brier 0.119±0.008. In the hold-out validation cohort (n=147; events=21), AUC and Brier were 0.499 and 0.127, respectively. Supplementary Table S5 provides full details. Calibration of the model was acceptable based on internal assessment, indicating agreement between predicted and observed risks. DCA (Fig. 3) indicated that the model provided superior net clinical benefit across a wide range of threshold probabilities, compared with the “treat-all” and “treat-none” strategies. A nomogram incorporating the five predictors was constructed (Fig. 4) to facilitate individualized clinical application. Using development-derived cut-points, the validation cohort showed the following operating characteristics: Youden 0.132 (sensitivity 0.48, specificity 0.57), rule-out 0.107 (sensitivity 0.52, specificity 0.48), and rule-in 0.191 (sensitivity 0.24, specificity 0.79). We propose low (<0.10), intermediate (0.10–0.19), and high (≥0.19) risk strata to guide prehabilitation intensity (Supplementary Table S6).
Subgroup analysis
Table 3 presents the model’s predictive performance across key clinical subgroups. In patients aged >65 years, the model achieved an AUC of 0.90 (95% CI 0.82–0.97), with a sensitivity and specificity of 0.87. Among those with a BMI<25 kg/m2, the model yielded an AUC of 0.84 (95% CI 0.75–0.94), sensitivity of 0.81, and specificity of 0.84. In the diabetic subgroup, the model demonstrated an AUC of 0.86 (95% CI 0.75–0.97), with a sensitivity of 0.82 and specificity of 0.92. These results indicate that the model maintained robust discriminatory capacity across patient subgroups with varying baseline risk profiles.
DISCUSSION
In this study, we identified five independent and modifiable predictors of PPCs in patients undergoing cardiac surgery: preoperative higher PaCO2, lower SaO2, slower walking speed, lower FEV1/FVC ratio and diminished inspiratory muscle strength (MIP % predicted). These findings formed the basis of a clinically applicable prediction model that demonstrated strong performance across both the overall cohort and key patient subgroups.
Compared with existing PPC prediction models, our approach offers several important improvements in both clinical relevance and practical application. Classical reference models were developed for broad surgical populations and, in the case of LAS VEGAS, require intraoperative variables (airway device, intraoperative desaturation, positive end-expiratory pressure levels, vasopressor use) [10]. The ARISCAT index (seven pre/intra-episode factors including preoperative SpO2, recent respiratory infection, anemia, surgical site, duration ≥2 hours, emergency surgery) showed excellent discrimination in its original cohorts (AUC 0.90/0.88) but was not designed specifically for cardiac surgery [5]. The NSQIP calculators predict single endpoints (pneumonia; respiratory failure) with preoperative factors such as ASA class, functional dependence, and sepsis [2,11]. Because our dataset lacks several required inputs (e.g., recent infection, procedure duration/emergency, ASA class, functional dependence, intraoperative ventilation metrics), a numerical head-to-head without variable re-specification was not feasible. We therefore provide a descriptive comparison (Supplementary Table S7) and avoid claims of superiority. Our model’s distinct contribution is to focus on modifiable, preoperative respiratory and functional markers in a cardiac-surgery population, thereby directly informing prehabilitation.
Elevated preoperative PaCO2 was found to be a significant risk factor for PPCs. Hypercapnia may reflect chronic hypoventilation or impaired pulmonary gas exchange and has been associated with increased risks of pneumonia, acute respiratory distress syndrome, and prolonged mechanical ventilation [20]. As reported by Kerr and Mills [21], preoperative hypercapnia may increase pulmonary vascular resistance and inflammation, compromise alveolar function, and thereby heighten susceptibility to postoperative lung complications. Building on its pathophysiology, a brief preoperative “optimization bundle” is pragmatic: step-up inhaled bronchodilators with verified device technique, daily airway-clearance drills, targeted screening for obstructive sleep apnea (OSA), with initiation of continuous positive airway pressure or noninvasive ventilation when indicated, and a short inspiratory-muscle-training block (~30% to 50% MIP, 15–20 minutes twice daily). Randomized and guideline evidence supports this pathway in cardiac surgery candidates [15,22].
Preoperative SpO2 breathing room air in supine position was the strongest patient-related PPC risk factor. We consider this to be a highly useful finding because SpO2 is an easily recorded objective measure. As demonstrated by Sampsonas et al. [23], SpO2<90% was associated with an increased risk of PPCs. Our model confirms this finding and extends it specifically to the cardiac surgical population, in which desaturation may coexist with subtle cardiopulmonary impairment not always captured by imaging or spirometry. We found a strong association between PPCs and respiratory disease (respiratory symptoms), smoking (lifetime exposure), and heart failure, consistent with previous studies [24]. However, these factors were not selected as independent predictors on multivariable analysis, probably because SpO2 is a reflection of both respiratory and cardiovascular functional status [5]. To translate gas-exchange risk into action, use ambulatory oximetry and titrate oxygen to guideline targets (typically 94%–98%, or 88%–92% when hypercapnia risk is present), promptly treat reversible contributors (bronchospasm, congestion), and start CPAP preoperatively when OSA is suspected/confirmed; simple diaphragmatic/pursed-lip breathing and daily airway-clearance practice can be delivered within 1–2 weeks to improve resting/exertional saturation [25-27].
Slow preoperative gait speed (≤0.83 m/s) strongly predicts PPCs after cardiac surgery, as it reflects weakened respiratory muscles, reduced heart-lung fitness, and poor inflammation control—key drivers of pneumonia and prolonged ventilator use [28]. Older adults or those with COPD/malnutrition face doubled PPC risks due to overlapping vulnerabilities [28]. Critically, improving gait speed or leg strength before surgery can cut PPC rates by 60%: programs combining breathing exercises (IMT) and leg resistance training boost both muscle power and lung function [29]. Consistent with the role of functional reserve, a condensed prehabilitation plan—moderate-intensity walking or cycling for 20–30 minutes on most days, sit-to-stand and step-ups for strength, and daily mobility targets—can be implemented even within 1–2 weeks; reviews in cardiac surgery associate such preoperative exercise (often alongside IMT) with lower PPCs and shorter length of stay (LOS), aligning with pulmonary rehabilitation and Enhanced Recovery After Surgery principles [27,30].
Preoperative obstructive ventilatory dysfunction, defined as FEV1/FVC, independently predicts PPCs after cardiac surgery. This threshold closely aligns with our training cohort data, where PPCs patients exhibited FEV1/FVC values of 72.93±9.73%. Previous studies confirmed that preoperative IMT reduces PPCs by 87% through enhanced respiratory strength and improved FEV1/FVC ratios [3,31,32]. High-intensity IMT further mitigates postoperative lung volume loss, while adjuncts like incentive spirometry or core stabilization training optimize ventilation efficiency [33]. These strategies collectively address respiratory muscle weakness and airflow limitation, positioning FEV1/FVC as both a prognostic marker and modifiable therapeutic target [3,34]. Clinically, airflow obstruction supports a short ‘prehab micro-cycle’: aerobic plus resistance training with airway-clearance skills over 2–4 weeks when feasible, alongside guideline-concordant escalation to dual long-acting bronchodilators (a long-acting beta-agonist combined with a long-acting muscarinic antagonist), with inhaled corticosteroid added when indicated, and strict checks of inhaler technique and adherence; reserve systemic steroids for true exacerbations [35]. Moreover, A focused IMT protocol (start ~30% MIP, progress to 40%–50%, 15–20 min twice daily for ≥1–2 weeks) paired with coached cough/huff techniques is practical and consistently linked to fewer PPCs and shorter LOS in cardiac surgery cohorts, including an randomized controlled trial (RCT) in high-risk CABG and subsequent systematic reviews [36].
Prolonged mechanical ventilation is a major contributor to PPCs in cardiac surgery patients. The need for extended ventilatory support often arises from impaired respiratory muscle performance, which itself may result from multiple surgical and physiological factors. For example, internal mammary artery harvesting during CABG reduces blood flow to the intercostal muscles, contributing to respiratory muscle weakness [37,38]. In addition, median sternotomy—a standard approach in cardiac surgery—causes significant chest wall disruption and pain, which, when combined with postoperative sedation, promotes shallow breathing and hypoventilation. Weakened respiratory muscles compromise both ventilatory efficiency and essential airway-protective actions such as coughing and swallowing. These impairments prolong ventilator dependence and intensive care unit stay, increasing the risk of ventilator-associated complications, impaired secretion clearance, and infection—all of which contribute to PPCs and adverse outcomes.
Importantly, the value of our model lies not only in discrimination but also in informing preoperative care. All five predictors identified—gas exchange abnormalities, reduced FEV1/FVC ratio, diminished inspiratory muscle strength, and slow gait speed—are clinically modifiable and align with established prehabilitation components. Preoperative IMT has been shown to significantly reduce PPC incidence in cardiac surgery patients, with several RCTs demonstrating lower complication rates after brief (5 day) IMT regimens [13]. Prehabilitation programs that integrate respiratory training with aerobic conditioning and nutrition support improve postoperative outcomes and functional recovery in this population [39]. Gait speed, a simple yet powerful marker of overall frailty, has been independently linked to mortality and PPC risk; slower preoperative gait is consistently associated with worse outcomes—even after adjusting for comorbidities [40]. However, as a single-center prospective observational study, our analyses estimate associations rather than causal effects; residual or unmeasured confounding, center-specific practices, and incomplete capture of perioperative management may remain. Accordingly, these suggestions should be considered decision-support and hypothesis-generating: the model may help prioritize candidates for targeted prehabilitation (e.g., IMT, aerobic exercise, nutritional optimization), but external validation and prospective impact studies—ideally randomized trials—are needed to determine whether acting on these markers reduces PPC risk and improves recovery.
Limitations
This study has several limitations. First, it was conducted at a single center with a relatively short enrollment period, which may limit the generalizability of the findings. Although internal validation was performed, external validation in multicenter cohorts is warranted to confirm the robustness of the model. Second, while the selected predictors are physiologically plausible and modifiable, the effects of targeted preoperative interventions on actual PPC incidence were not tested in this observational study. Third, the diagnosis of PPCs relied on clinical judgment and the Kroenke score, which may introduce interobserver variability. Lastly, some potentially relevant variables—such as frailty scales, detailed pulmonary imaging findings, or long-term outcomes—were not included and merit future exploration.
Conclusion
We developed a clinically actionable model to predict PPCs in cardiac surgery patients, based on five modifiable preoperative factors: gas exchange abnormalities, reduced FEV1/FVC ratio, diminished inspiratory muscle strength, and slow gait speed. The model showed strong predictive accuracy and supports targeted prehabilitation strategies. By linking risk stratification with intervention, it provides a practical tool to improve perioperative outcomes.
CONFLICTS OF INTEREST
No potential conflict of interest relevant to this article was reported.
FUNDING INFORMATION
This research was supported by the National High Level Hospital Clinical Research Funding (Grant No.2023-GSP-QN-38 & Grant No.2023-GSP-RC-07) and the National Natural Science Foundation of China (Grant No. 82370401).
AUTHOR CONTRIBUTION
Conceptualization: Li R, Tian M, Wang C, Liu B, Feng X. Data curation: Li R, Huang Y, Chen W, Song Y, Liu B. Formal analysis: Li R, Tian M. Funding acquisition: Li R, Tian M, Feng X. Investigation: Li R, Tian M. Methodology: Li R, Tian M, Wang C, Chen W, Song Y, Liu B, Feng X. Project administration: Du L, Feng X. Supervision: Du L, Feng X. Software: Li R, Tian M, Chen W. Validation: Li R, Song Y, Liu B. Writing – original draft: Li R. Writing – review & editing: Li R, Tian M, Wang C. Approval of final manuscript: all authors.
Receiver operating characteristic curves for the prediction model in the training and validation cohorts. The model showed good discrimination with an AUC of 0.86 (95% CI 0.80–0.93) in the training cohort and 0.87 (95% CI 0.74–0.93) in the validation cohort. AUC, area under the receiver operating characteristic curve; CI, confidence interval.
Fig. 3.
DCA of the prediction model. The DCA curve illustrates that the model provides greater net clinical benefit across a range of threshold probabilities, compared to the “treat-all” and “treat-none” strategies, supporting its value in clinical decision-making. DCA, decision curve analysis.
Fig. 4.
Nomogram for predicting PPCs after cardiac surgery. The nomogram was constructed based on five independent predictors identified in the multivariable logistic regression model: arterial PaCO2, oxygen saturation<93% (SpO2), gait speed, FEV1/FVC ratio, and inspiratory muscle strength (% predicted). Each predictor corresponds to a point score, and the total score estimates the individual probability of developing PPCs. This visual tool is intended to support individualized preoperative risk stratification. For bedside use, illustrative clinical thresholds are 10% (rule-out-oriented) and 19% (rule-in-oriented); operating characteristics and risk strata are provided in Supplementary Table S6. PPCs, postoperative pulmonary complications; PaCO2, partial pressure of arterial carbon dioxide; SpO2, peripheral capillary oxygen saturation; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; MIP, maximal inspiratory pressure.
Table 1.
Baseline characteristics of patients with and without PPCs in the training cohort
b)PaO2 refer to arterial partial pressures of carbon dioxide and oxygen, respectively, measured on hospital admission as part of routine preoperative assessment.
c)Preoperative anemia was defined as hemoglobin <130 g/L in men or <120 g/L in women.
Discriminatory performance of the PPC prediction model in the training and validation cohorts
Cohort
AUC (95% CI)
Sensitivity
Specificity
Training cohort
0.86 (0.80–0.93)
0.78
0.91
Validation cohort
0.87 (0.74–0.93)
0.81
0.87
PPCs, postoperative pulmonary complications; AUC, area under the receiver operating characteristic curve; CI, confidence interval.
REFERENCES
1. Hadaya J, Verma A, Marzban M, Sanaiha Y, Shemin RJ, Benharash P. Impact of pulmonary complications on outcomes and resource use after elective cardiac surgery. Ann Surg 2023;278:e661-6.
2. Gupta H, Gupta PK, Schuller D, Fang X, Miller WJ, Modrykamien A, et al. Development and validation of a risk calculator for predicting postoperative pneumonia. Mayo Clin Proc 2013;88:1241-9.
3. Cursino de Moura JF, Oliveira CB, Coelho Figueira Freire AP, Elkins MR, Pacagnelli FL. Preoperative respiratory muscle training reduces the risk of pulmonary complications and the length of hospital stay after cardiac surgery: a systematic review. J Physiother 2024;70:16-24.
6. Zochios V, Klein AA, Gao F. Protective invasive ventilation in cardiac surgery: a systematic review with a focus on acute lung injury in adult cardiac surgical patients. J Cardiothorac Vasc Anesth 2018;32:1922-36.
7. Ibañez J, Riera M, Amezaga R, Herrero J, Colomar A, Campillo-Artero C, et al. Long-term mortality after pneumonia in cardiac surgery patients: a propensity-matched analysis. J Intensive Care Med 2016;31:34-40.
8. Fernandez-Bustamante A, Frendl G, Sprung J, Kor DJ, Subramaniam B, Martinez Ruiz R, et al. Postoperative pulmonary complications, early mortality, and hospital stay following noncardiothoracic surgery: a multicenter study by the perioperative research network investigators. JAMA Surg 2017;152:157-66.
9. Mazo V, Sabaté S, Canet J, Gallart L, de Abreu MG, Belda J, et al. Prospective external validation of a predictive score for postoperative pulmonary complications. Anesthesiology 2014;121:219-31.
10. Neto AS, da Costa LGV, Hemmes SNT, Canet J, Hedenstierna G, Jaber S, et al.; LAS VEGAS. The LAS VEGAS risk score for prediction of postoperative pulmonary complications: an observational study. Eur J Anaesthesiol 2018;35:691-701.
11. Gupta H, Gupta PK, Fang X, Miller WJ, Cemaj S, Forse RA, et al. Development and validation of a risk calculator predicting postoperative respiratory failure. Chest 2011;140:1207-15.
13. Chen X, Hou L, Zhang Y, Liu X, Shao B, Yuan B, et al. The effects of five days of intensive preoperative inspiratory muscle training on postoperative complications and outcome in patients having cardiac surgery: a randomized controlled trial. Clin Rehabil 2019;33:913-22.
14. Kroenke K, Lawrence VA, Theroux JF, Tuley MR. Operative risk in patients with severe obstructive pulmonary disease. Arch Intern Med 1992;152:967-71.
15. Hulzebos EH, Helders PJ, Favié NJ, De Bie RA, Brutel de la Riviere A, et al. Preoperative intensive inspiratory muscle training to prevent postoperative pulmonary complications in high-risk patients undergoing CABG surgery: a randomized clinical trial. JAMA 2006;296:1851-7.
16. Zhang J, Zhang H, Du J, Feng W. Comparative study on early and long-term outcomes between on-pump and off-pump coronary artery bypass grafting in elder female patients. Chin Circ J 2015;12:1143-6.
17. Grønkjær M, Eliasen M, Skov-Ettrup LS, Tolstrup JS, Christiansen AH, Mikkelsen SS, et al. Preoperative smoking status and postoperative complications: a systematic review and meta-analysis. Ann Surg 2014;259:52-71.
18. Covarrubias J, Grigorian A, Schubl S, Gambhir S, Dolich M, Lekawa M, et al. Obesity associated with increased postoperative pulmonary complications and mortality after trauma laparotomy. Eur J Trauma Emerg Surg 2021;47:1561-8.
19. Sun HN, Hu SS, Zheng Z, Hou JF. The influence of smoking on early outcomes in patients with coronary artery bypass graft surgery. Chin Circ J 2011;26:89-92. Chinese.
21. Kerr K, Mills GH. Intra-operative and post-operative hypercapnia leading to delayed respiratory failure associated with transanal endoscopic microsurgery under general anaesthesia. Br J Anaesth 2001;86:586-9.
22. Chung F, Memtsoudis SG, Ramachandran SK, Nagappa M, Opperer M, Cozowicz C, et al. Society of anesthesia and sleep medicine guidelines on preoperative screening and assessment of adult patients with obstructive sleep apnea. Anesth Analg 2016;123:452-73.
23. Sampsonas F, Antonogiannaki M, Vittorakis S, Kyriakopoulos C, Boutlas S, Boutou A, et al. Preoperative evaluation of the respiratory system: a narrative review based on Hellenic Thoracic Society guidelines. Pneumon 2023;36:18.
24. Smetana GW, Lawrence VA, Cornell JE; American College of Physicians. Preoperative pulmonary risk stratification for noncardiothoracic surgery: systematic review for the American College of Physicians. Ann Intern Med 2006;144:581-95.
25. O'Driscoll BR, Howard LS, Earis J, Mak V; British Thoracic Society Emergency Oxygen Guideline Group; BTS Emergency Oxygen Guideline Development Group. BTS guideline for oxygen use in adults in healthcare and emergency settings. Thorax 2017;72(Suppl 1):ii1-90.
26. American Society of Anesthesiologists Task Force on Perioperative Management of patients with obstructive sleep apnea. Practice guidelines for the perioperative management of patients with obstructive sleep apnea: an updated report by the American Society of Anesthesiologists Task Force on Perioperative Management of patients with obstructive sleep apnea. Anesthesiology 2014;120:268-86.
27. Spruit MA, Singh SJ, Garvey C, ZuWallack R, Nici L, Rochester C, et al.; ATS/ERS Task Force on Pulmonary Rehabilitation. An official American Thoracic Society/European Respiratory Society statement: key concepts and advances in pulmonary rehabilitation. Am J Respir Crit Care Med 2013;188:e13-64.
28. Chang J, Nathalie J, Nguyenhuy M, Xu R, Virk SA, Saxena A. Slow gait speed is associated with worse postoperative outcomes in cardiac surgery: a systematic review and meta-analysis. J Card Surg 2022;37:197-204.
29. Xiang Y, Zhao Q, Luo T, Zeng L. Inspiratory muscle training to reduce risk of pulmonary complications after coronary artery bypass grafting: a systematic review and meta-analysis. Front Cardiovasc Med 2023;10:1223619.
31. Silva HVC, Lunardi AC, Pinto ACPN, Macedo JRFF, Santos EDCD. Is incentive spirometry superior to standard care in postoperative cardiac surgery on clinical outcomes and length of hospital and intensive care unit stay? A systematic review with meta-analysis. Braz J Cardiovasc Surg 2024;39:e20220319.
32. Wang J, Wang YQ, Shi J, Yu PM, Guo YQ. Effect of preoperative inspiratory muscle training on postoperative outcomes in patients undergoing cardiac surgery: a systematic review and meta-analysis. World J Clin Cases 2023;11:2981-91.
33. Su H, Zhang J, Liu Y, Peng H, Zhang L. Pre and postoperative nurse-guided incentive spirometry versus physiotherapist-guided pre and postoperative breathing exercises in patients undergoing cardiac surgery: an evaluation of postoperative complications and length of hospital stay. Medicine (Baltimore) 2022;101:e32443.
34. Furon Y, Dang Van S, Blanchard S, Saulnier P, Baufreton C. Effects of high-intensity inspiratory muscle training on systemic inflammatory response in cardiac surgery - a randomized clinical trial. Physiother Theory Pract 2024;40:778-88.
36. Katsura M, Kuriyama A, Takeshima T, Fukuhara S, Furukawa TA. Preoperative inspiratory muscle training for postoperative pulmonary complications in adults undergoing cardiac and major abdominal surgery. Cochrane Database Syst Rev 2015;2015:CD010356.
37. Baumgarten MC, Garcia GK, Frantzeski MH, Giacomazzi CM, Lagni VB, Dias AS, et al. Pain and pulmonary function in patients submitted to heart surgery via sternotomy. Rev Bras Cir Cardiovasc 2009;24:497-505. English, Portuguese.
38. Praveen R, Swaminathan N, Praveen JS. Inspiratory muscle training is effective in improving respiratory muscle functions in patients who have undergone coronary artery bypass graft. Fizjoterapia Polska 2009;9:285-92.
39. Cordeiro ALL, Soares LO, Gomes-Neto M, Petto J. Inspiratory muscle training in patients in the postoperative phase of cardiac surgery: a systematic review and meta-analysis. Ann Rehabil Med 2023;47:162-72.
40. Afilalo J, Sharma A, Zhang S, Brennan JM, Edwards FH, Mack MJ, et al. Gait speed and 1-year mortality following cardiac surgery: a landmark analysis from the society of thoracic surgeons adult cardiac surgery database. J Am Heart Assoc 2018;7:e010139.
Development and Validation of a Clinically Actionable Prediction Model for Postoperative Pulmonary Complications in Cardiac Surgery: A Focus on Modifiable Risk Factors
Fig. 1. Flow diagram of the study.
Fig. 2. Receiver operating characteristic curves for the prediction model in the training and validation cohorts. The model showed good discrimination with an AUC of 0.86 (95% CI 0.80–0.93) in the training cohort and 0.87 (95% CI 0.74–0.93) in the validation cohort. AUC, area under the receiver operating characteristic curve; CI, confidence interval.
Fig. 3. DCA of the prediction model. The DCA curve illustrates that the model provides greater net clinical benefit across a range of threshold probabilities, compared to the “treat-all” and “treat-none” strategies, supporting its value in clinical decision-making. DCA, decision curve analysis.
Fig. 4. Nomogram for predicting PPCs after cardiac surgery. The nomogram was constructed based on five independent predictors identified in the multivariable logistic regression model: arterial PaCO2, oxygen saturation<93% (SpO2), gait speed, FEV1/FVC ratio, and inspiratory muscle strength (% predicted). Each predictor corresponds to a point score, and the total score estimates the individual probability of developing PPCs. This visual tool is intended to support individualized preoperative risk stratification. For bedside use, illustrative clinical thresholds are 10% (rule-out-oriented) and 19% (rule-in-oriented); operating characteristics and risk strata are provided in Supplementary Table S6. PPCs, postoperative pulmonary complications; PaCO2, partial pressure of arterial carbon dioxide; SpO2, peripheral capillary oxygen saturation; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; MIP, maximal inspiratory pressure.
Graphical abstract
Fig. 1.
Fig. 2.
Fig. 3.
Fig. 4.
Graphical abstract
Development and Validation of a Clinically Actionable Prediction Model for Postoperative Pulmonary Complications in Cardiac Surgery: A Focus on Modifiable Risk Factors
Characteristic
Training cohort
p-value
No PPCs (n=296)
PPCs (n=49)
Sex, female
88.0 (29.7)
13.0 (26.5)
0.78
Age (yr)
59.0±11.1
61.7±9.2
0.11
Body mass index (kg/m2)
25.4±3.3
25.4±3.8
0.94
Prealbumin (mg/L)
245.3±53.2
250.7±52.3
0.51
Total protein (g/L)
65.9±5.5
65.3±4.4
0.53
Albumin (g/L)
41.3±3.4
41.4±3.0
0.78
LDL-C (mmol/L)
2.2±0.8
2.1±0.7
0.64
Ferritin (µg/L)
189.1±134.0
182.7±114.2
0.75
PaCO2a) (mmHg)
38.3±3.5
41.9±5.0
<0.001
PaO2b) (mmHg)
89.4±10.9
65.5±23.0
<0.001
Pre-op SpO2<93%
12 (4.1)
30 (61.2)
<0.001
Preoperative anemiac)
11 (3.7)
3 (6.1)
0.689
Diabetes mellitus
87 (29.4)
11 (22.4)
0.41
COPD
9 (3.0)
1 (2.0)
>0.999
History of smoking
124 (41.9)
20 (40.8)
>0.999
Current smoker
43 (14.5)
6 (12.2)
0.84
Grip strength indexd) (%)
45.7±22.9
42.4±11.8
0.32
Gait speede) (m/s)
1.3 (1.0–1.6)
1.2 (0.9–1.6)
0.02
Chair rise time (s)
10.4±3.2
11.4±2.2
0.04
FVC (L)
3.1±0.9
3.1±0.8
0.83
PEF (L/s)
5.8±1.8
5.6±1.9
0.50
FEV1 (L)
2.38±0.7
2.2±0.7
0.15
FEV1/FVC ratio (%)
78.0±9.3
72.9±9.7
0.01
MVV (L/min)
69.5±28.8
68.0±30.0
0.75
MIPf) (cmH2O)
66.6±19.5
61.4±21.1
0.09
MIP (% of predicted)
69.6±15.9
63.2±19.2
0.01
MEPg) (cmH2O)
101.3±35.4
97.3±35.8
0.46
MEP (% of predicted)
82.1±23.5
77.9±26.7
0.26
Characteristic
Univariable
Multivariable
OR (95% CI)
p-value
OR (95% CI)
p-value
Sex, female
0.85 (0.42–1.65)
0.65
Age
1.03 (1.00–1.06)
0.11
PaCO2
1.30 (1.19–1.44)
<0.001
1.12 (1.00–1.26)
0.04
PaO2
0.92 (0.89–0.94)
<0.001
Pre-op SpO2<93%
37.37 (17.01–87.52)
<0.001
12.47 (3.51–48.13)
<0.001
Gait speed
0.20 (0.08–0.48)
<0.001
0.17 (0.04–0.71)
0.03
Chair rise time
1.10 (1.00–1.21)
0.04
FEV1/FVC ratio
0.95 (0.92–0.98)
0.001
0.96 (0.92–1.00)
0.04
MIP
0.99 (0.97–1.00)
0.09
MIP (% of predicted)
0.97 (0.95–0.99)
0.01
0.96 (0.92–0.99)
0.02
Cohort
AUC (95% CI)
Sensitivity
Specificity
Training cohort
0.86 (0.80–0.93)
0.78
0.91
Validation cohort
0.87 (0.74–0.93)
0.81
0.87
Table 1. Baseline characteristics of patients with and without PPCs in the training cohort
Values are presented as number (%) or mean±standard deviation.
PaO2 refer to arterial partial pressures of carbon dioxide and oxygen, respectively, measured on hospital admission as part of routine preoperative assessment.
Preoperative anemia was defined as hemoglobin <130 g/L in men or <120 g/L in women.
, Meice Tian, MD2,*
